Delaunay triangulation divides a set of points into triangles so that no point lies inside any triangle’s circumcircle. This rule avoids narrow, sliver-shaped triangles by maximizing the minimum angle. section 5.4.1
As a prediction task, it asks a Ptr-Net to choose the triangles connecting a given set of points. Classical computational geometry algorithms provide exact solutions for comparison. The network’s full reconstruction weakened as point sets grew, even when it still predicted many individual triangles correctly. section 5.4.1
Accuracy and triangle coverage
With 5 points, the network achieved 80.7% accuracy and 93.0% triangle coverage. With 50 points, it achieved no perfect reconstructions but still covered 52.8% of the triangles. section 5.4.1
